---
title: "What can be local and what needs a connection"
description: "The raw speech-to-text route is local once its required components are ready."
slug: "offline-scope"
category: "privacy-data"
language: en
editorial_status: published
capability_status: available
reviewed_at: 2026-10-13
keywords: ["CueSpec offline transcription","local AI voice input","network dependencies"]
---

# What can be local and what needs a connection

The raw speech-to-text route is local once its required components are ready. Updates, model preparation, optional refinement and Codex task execution have separate connectivity requirements.

## Separate the dependencies
A downloaded and ready speech model supports local recognition. Optional text refinement requires its configured service and login. Sending and following an AI task depends on the host task environment rather than the transcription engine alone.

If the network is unavailable, a locally produced draft may still be useful, but that does not establish successful task delivery. Retain the draft and verify receipt when the environment is available.

## Test only what you need
To assess local recognition, use a non-sensitive sentence and confirm draft creation. To assess sending, check the original conversation. To assess refinement, compare raw and revised text with the service actually active.

Do not summarize the entire product as “fully offline” from one local stage. Equally, a service-refinement failure should not imply that raw transcription requires that service. Clear stage-specific documentation helps you choose the mode appropriate for the task and understand what a connection failure affects.

## Related reading

- [Prepare local voice input](/prepare-local-voice/)
- [Where recorded audio goes](/audio-data-flow/)
- [Local processing does not prevent overhearing](/public-space-confidentiality/)
- [CueSpec product facts at a glance](/product-facts/)